SOX, Corporate Transparency, and the Cost of Debt

SOX, Corporate Transparency, and the Cost of Debt Sandro C. Andrade University of Miami

Gennaro Bernile University of Miami and SEC

Frederick M. Hood Virginia Tech

First version: August 11, 2008. This version: March 1, 2009.

Abstract We investigate the impact of the Sarbanes-Oxley (SOX) Act on the cost of debt through its e¤ect on the reliability of …nancial reporting. Using Credit Default Swap (CDS) spreads and a structural CDS pricing model, we calibrate a …rm-level corporate opacity parameter in the preand post-SOX periods. Our analysis shows that corporate opacity and the cost of debt decrease signi…cantly after SOX. The median …rm in our sample experiences an 18 bp reduction on its …veyear CDS spread as a result of lower opacity following SOX, amounting to total annual savings of $ 844 million for the 252 …rms in our sample. Furthermore, the reduction in opacity tends to be larger for …rms that in the pre-SOX period have lower accrual quality, less conservative earnings, lower number of independent directors, lower S&P Transparency and Disclosure ratings, and are more likely to bene…t from SOX-compliance according to Chhaochharia and Grinstein’s (2007) criteria.

Keywords: Sarbanes-Oxley, Corporate transparency, CDS pricing. JEL number: G38, G33, G12

We are grateful to Dan Bergstresser, Je¤ Bohn, Doug Emery, Gregg Jarrell, Kathleen Hanley, Qiang Kang, Andy Leone, DJ Nanda, Josh White, Peter Wysocki, Jerry Zimmermann, an anonymous referee, and to Vidhi Chhaochharia especially. We bene…ted from useful comments by seminar participants at American University, University of Miami, Virginia Tech, Securities Exchange Comission, and the Australasian Finance and Banking Conference. All errors are ours. E-mail addresses: [email protected], [email protected] and [email protected].

1

SOX, Corporate Transparency, and the Cost of Debt

Abstract We investigate the impact of the Sarbanes-Oxley (SOX) Act on the cost of debt through its e¤ect on the reliability of …nancial reporting. Using Credit Default Swap (CDS) spreads and a structural CDS pricing model, we calibrate a …rm-level corporate opacity parameter in the preand post-SOX periods. Our analysis shows that corporate opacity and the cost of debt decrease signi…cantly after SOX. The median …rm in our sample experiences an 18 bp reduction on its …veyear CDS spread as a result of lower opacity following SOX, amounting to total annual savings of $ 844 million for the 252 …rms in our sample. Furthermore, the reduction in opacity tends to be larger for …rms that in the pre-SOX period have lower accrual quality, less conservative earnings, lower number of independent directors, lower S&P Transparency and Disclosure ratings, and are more likely to bene…t from SOX-compliance according to Chhaochharia and Grinstein’s (2007) criteria.

Keywords: Sarbanes-Oxley, Corporate transparency, CDS pricing. JEL number: G38, G33, G12

1

1

Introduction

The enactment of the Sarbanes-Oxley (SOX) Act in July 2002 is arguably one of the most signi…cant regulatory events in the recent history of US capital markets. Advocates of the Act claim its main objective was to “rebuild public trust in US capital markets”, whose functioning had been undermined by the deteriorating quality of accounting information culminating in a series of accounting scandals (Cohen, Dey, and Lys, 2008; Jorion, Shi, and Zhang, 2007; Healy and Palepu, 2003). To that end, the Act contains several mandates aiming to increase corporate transparency through more reliable corporate reporting. According to Coates (2007), the two core components of such mandates are the creation of a quasi-public institution to supervise auditors, and the enlisting of auditors to enforce new disclosure rules giving …rms incentives to tighten …nancial controls. The Sarbanes-Oxley Act imposes both direct and indirect costs on public …rms. Direct out-ofpocket costs include internal compliance costs and increased audit fees (Illiev 2007), while indirect costs arise from sub-optimal disclosure under tighter constraints compared to laxer ones (Verrechhia 1983). The indirect costs of excessive disclosure may include competitive disadvantages in product markets; bargaining disadvantages with customers, suppliers, and employees; and increased risk exposure of top o¢ cers resulting in risk avoiding behavior (Hermalin and Weisbach 2007; Bargeron, Lehn, and Zutter 2008; Kang, Liu, and Qi 2008). The net bene…ts of the new legislation, if any, are still under debate1 . In this paper we focus on an aspect of SOX that has received little attention: the e¤ect of the Act on the cost of debt capital due to presumably higher reliability of corporate reporting. Admittedly we do not provide a full cost-bene…t analysis of the Act; instead we attempt to shed light on a particular e¤ect of the legislation that is arguably hard to measure. Our results show a median decrease in the cost of debt of 17:7 basis points per year for our sample …rms due to an increase in corporate transparency as perceived by investors. This e¤ect is economically large, considering that the risk-free rate and the median credit spread were respectively 330 and 111 basis points in the period immediately after the passage of the Act. In dollar terms, the perceived improvement in the quality of …nancial reporting translates into total savings of US$ 843 million per year for the 252 …rms in our sample. Consistent with previous studies, our evidence indicates that the e¤ect of the Act depends on predictable …rms characteristics (Chhaochharia and Grinstein 2007; Zhang 2008). Speci…cally, the reduction in opacity perceived by investors following SOX is larger for …rms that: are less transparent according to the 2002 S&P Transparency and Disclosure Index, have lower earnings quality in the pre-SOX period, have lower number of independent directors, and are 1

See Bushee and Leuz (2005); Chhaochharia and Grinstein (2007); Zhang (2007); Leuz (2007); Iliev (2007); Hostak

et al. (2008); Ashbaugh-Skaife et al. (2008), for analyses of the economic consequences of SOX.

2

more likely to be a¤ected by SOX according to the criteria in Chhaochharia and Grinstein (2007). Perhaps the large e¤ect of SOX on credit spreads we document is not surprising: recent research underscores the importance of corporate transparency for the pricing of debt-related contracts. In their in‡uential work, Du¢ e and Lando (2001) develop a model showing that corporations with less reliable …nancial reports have higher secondary market credit spreads due to the asymmetric nature of cash ‡ows from debt contracts. This occurs even when investors are risk-neutral and symmetrically informed. The Du¢ e-Lando model is able to generate non-negligible short-term credit spreads for investment grade corporations, a robust empirical phenomenon that is hard to explain in a full information framework. Empirical research by Sengupta (1998), Anderson, Mansi, and Reeb (2004), Mansi, Maxwell, and Miller (2004), Yu (2005), Ball, Robin, and Sadka (2008), Zhang (2008), Wittenberg-Moerman (2008), and Duarte, Young, and Yu (2008) corroborates the importance of corporate transparency for debt pricing. A contemporaneous and independent paper by DeFond, Hung, Karaoglu, and Zhang (2008) also studies the impact of SOX on debt prices. Using cumulative "abnormal" changes in corporate bond spreads over 13 short-term windows surrounding events leading up to the passage of SOX, they conclude that the Act increased the cost of debt by 20 basis points. Our work di¤ers from theirs in at least three important ways. First, in the same spirit of Chhaochharia and Grinstein (2007), we use long pre- and post-SOX windows rather than price changes over few days around selected pre-enactment events2 . Second, our analysis relies on CDS spreads not corporate bond prices. The secondary market for corporate bonds is less liquid and with larger bid–ask spreads than the CDS market, which may pose a challenge for event spreads study analyses, particularly those with short event windows such as DeFond et al. (2008)3 . Third, we rely on spread levels and a structural pricing model to calibrate …rm-period speci…c opacity parameters, and use the latter to evaluate the e¤ect of SOX on the cost of debt through its e¤ect on the reliability of corporate reports4 . In contrast, DeFond et al. use OLS regressions to detect "abnormal" changes in spreads. In the next Section we argue that non-linearities, interaction terms, and endogeneity cast doubt on the use of OLS regressions to address the e¤ect of SOX on credit spreads. 2

Both Chhaochharia and Grinstein (2007) and Zhang (2007) study the e¤ect of SOX on …rm value. Using

short-term event windows surrounding events leading up to the passage of SOX, Zhang (2007) concludes that, in value-weighted aggregation, SOX reduced …rms’ value. Chhaochharia and Grinstein (2007) use long term windows and reach the opposite conclusion. 3 Ait-Sahalia, Mykland and Zhang (2005) show that in markets whose prices are a¤ected by microstructure noise, short-window price variations are more a¤ected by noise than longer-window price variations. 4 Ball, Bushman, and Vasvari (2008) propose a measure of accounting quality based on the goodness-of-…t of a model of credit rating changes as a function of lagged earnings. In contrast, our opacity parameter is calibrated from the levels of CDS spreads and current market and accounting information.

3

The rest of the paper is organized as follows: in Section 2, we describe our methodology and data, and develop three hypotheses whose empirical tests are reported in Section 3. In this section, we also estimate the e¤ect of SOX due to increased reliability of corporate …nancial reporting, the main goal of the paper. In Section 4, we show our results are robust to plausible alternative explanations of our main …ndings and to sensible variations in our calibration procedure. Section 5 concludes the paper.

2

Methodology, Data, and Testable Hypotheses

We measure the cost of debt using credit spreads from Credit Default Swap (CDS) contracts. A CDS is an over-the-counter insurance contract on debt. The buyer and seller of insurance agree on a reference corporate bond and on a notional value for the contract, for example US$ 10 million. The buyer of insurance pays the quoted spread times $10 million to the seller of insurance, typically on a quarterly basis, and obtains the right to sell bonds with face value $10 million, at their face value, to the seller of insurance in the event of corporate default. CDS and corporate bond spreads are closely related theoretically and empirically (Du¢ e 1999; Blanco et al. 2005), but there are several advantages in using CDS rather than bond spreads in our research. First, CDS spreads are quoted directly, as opposed to bond spreads that depend on the arbitrary choice of a default-free term structure of interest rates. Second, traded CDS spreads have a …xed maturity, so it is not necessary to control for changes in time to maturity. Third, the CDS market has become much more liquid than the secondary market for corporate bonds, therefore CDS market prices are in principle more reliable (Hull et al. 2004; Blanco et al. 2005). Finally, in contrast to corporate bonds, there is no reason to believe that illiquidity in the CDS market a¤ects the average level of a …rm’s CDS spread because a CDS is a derivative contract not an asset (Longsta¤ et al. 2005).

2.1

CDS pricing model

Corporate transparency is only one of several determinants of credit spreads. In order to measure the change in spreads due a change in corporate reporting reliability, we need to control for changes in the other spread determinants. Controlling for other spread determinants using OLS regressions could lead to misspeci…cation because of non-linearities and interaction terms, and because of an important endogeneity issue. First, structural debt pricing models indicate that the derivative of credit spreads with respect to a 4

given spread determinant depends crucially on the level of that determinant and of other determinants. In other words, the impact of credit spread determinants is highly non-linear and includes important interactions among the determinants. Empirical research by Schaefer and Strebulaev (2008) con…rms that non-linearities and interactions are economically signi…cant. The authors show that the sensitivity of credit spreads to leverage is much higher at high spread levels than at low spread levels. Therefore, regressions of credit spreads would have to group …rms by spread levels at the minimum. Ideally, the regression speci…cation would include numerous powers and cross-products of explanatory variables. Second, …rms with less reliable corporate reporting, recognizing that they are charged relatively high interest rates, may choose to take on less debt. Therefore, if corporate opacity is imperfectly measured with existing proxies, OLS regressions of credit spreads on leverage and other explanatory variables yield biased and inconsistent coe¢ cient estimates because the residual is correlated with explanatory variables. Research by Molina (2005) indicates that the endogeneity of leverage is more than a mere technicality: accounting for it increases the e¤ect of leverage on default probabilities by a factor of three5 . Analogously, the endogeneity of leverage should matter for the relation between credit spreads and leverage. We address these empirical di¢ culties by using a structural debt pricing model that explicitly incorporates the e¤ect of accounting reliability, along with all the other credit spread determinants. We rely on the CreditGrades model, which delivers a simple, analytical debt pricing formula. The model was jointly developed by Goldman Sachs, JP Morgan and Deutsche Bank and is a popular debt pricing tool among practitioners. According to Currie and Morris (2002), the CreditGrades model was the industry standard CDS pricing model as of 2002. Attesting to the popularity of the model, Yu (2006) and Duarte, Longsta¤, and Yu (2007) use the CreditGrades model in recent research. In contrast to models of debt pricing under full information, the CreditGrades model explicitly incorporates a parameter representing uncertainty about the true level of a …rm’s liabilities. The logic underlying this extension is that the level of liabilities reported on the …rm’s balance sheet is potentially di¤erent from the level of liabilities that will drive a corporation to default. We refer to this uncertainty parameter as "corporate opacity". Our research strategy is to calibrate 5

Molina (2005) attributes the leverage endogeneity problem to imperfect measurement of fundamental risk: equity

or asset volatility would be imperfect proxies of fundamental business risk, therefore OLS regressions that attempt to control for fundamental risk by adding volatility as an explanatory variable (along with leverage) would yield biased and inconsistent coe¢ cients. Our point about the imperfect measurement of corporate transparency provides additional motivation for the leverage endogeneity problem. Molina (2005) uses IV estimation to circumvent the leverage endogeneity problem, using the history of …rms’ past market valuations and …rms’ marginal tax rates as intruments for the e¤ect of leverage on default probabilities.

5

this parameter for each …rm in the pre- and post-SOX periods by minimizing the sum of squared di¤erences between market and model-implied prices. By using …rm-level changes in calibrated corporate opacity in the pre- and post-SOX periods, we control for all the other credit spread determinants in the model taking into account interactions between them and non-linear e¤ects6 .

2.2

CDS pricing formula

The CreditGrades CDS pricing model requires eight inputs: time to expiration T ; stock price S; equity volatility

S;

recovery rate R; risk-free rate r; reported liabilities per equity share D;

expected location of the default boundary as fraction of liabilities L; and a parameter uncertainty about the location of the default boundary. Formally,

representing

is the standard deviation of

the log of the default boundary as a fraction of liabilities. We interpret

as a measure of corporate

opacity because when reported liabilities are less reliable there is more uncertainty about the true level of liabilities that will drive the …rm to default. The CreditGrades manual (2002) shows that the CDS spread can be well approximated by: c(T ) = r(1

R)

1

q(0) + H(T ) q(T )e rT H(T )

q(0)

(1)

The function q( ) is de…ned as q(t) = where

A(t) ln(d) + 2 2

A(t) 2

d

ln(d) A(t)

;

(2)

( ) is the standard normal c.d.f. and d=

S + LD 2 e ; LD

A(t) =

Finally,

p

2t

2

+

H(T ) = er (G(T + ) where 1

G(t) = dz+ 2

ln(d) p t

z

p

t +d

;

=

S

S : S + LD

G( )) ; z+ 12

(3) p ln(d) p +z t ; t

(4)

and 2

= z = 6

2

1 2r + 2: 4

(5)

A similar approach to account for non-linearity and interactions among credit spread determinants is used by

Davydenko and Strebulaev (2007). To study the e¤ect of strategic interactions between shareholders and debtholders on credit spreads, while taking account of other spread determinants, the authors compute the di¤erence between actual spreads and spreads implied by a structural debt pricing model without such strategic interactions. Then they regress these residuals onto theoretically motivated variables that might explain strategic interactions.

6

2.3

Data sources and sample selection

Using daily CDS quotes, we calibrate a corporate opacity parameter

for each …rm by minimizing

the sum of squared di¤erences between market CDS spreads and model-implied CDS spreads. We calibrate separate parameters before and after the enactment of SOX for each …rm in the sample. We de…ne the pre-SOX period as January 1, 2001 to July 31, 2002, and the post-SOX period as August 1, 2002 to December 31, 2003. To perform the calibrations, we require each …rm in the sample to have at least 30 CDS quotes in the pre-SOX period and 30 CDS quotes in the post-SOX period. We restrict the sample to non-…nancial …rms and main entities, as opposed to subsidiaries. Markit Partners provided us with the CDS data7 . Markit collects OTC dealer quotes on di¤erent CDS tenors on a daily basis. Until recently, volume in the CDS market was concentrated in 5-year contracts. Since we want liquid market quotes in our model calibration, we focus on the 5-year contract, as do other researchers. Also following the literature, we focus on US dollar denominated senior unsecured CDS contracts with the modi…ed restructuring clause (e.g. Jorion and Zhang 2007)8 . In addition to the corporate opacity parameter

, there are seven additional inputs required to

price the CDS as shown by Equations (1) to (6). The time to expiration is …xed at T =5 years. The stock price S is the common stock closing prices from CRSP. Following Hull, Predescu and White (2004), the risk-free rate r is the 5-year swap rate minus 10 basis points. Liabilities per share D is total liabilities minus minority interest and deferred taxes divided by the number of shares outstanding. Balance sheet information is from COMPUSTAT, based on the most recent annual statement available to investors at the time the market prices are quoted. The recovery rate R is from the Markit database, following Zhang, Zhou, and Zhu (2008). Along with CDS quotes, Markit also collects a daily …rm-speci…c estimate of the recovery value on a defaulted bond referenced by the CDS contract, provided by the quoting CDS dealers. Equity volatility

S

is the

…ve-year forecast from a GARCH(1,1) model …t on the full sample period, following Engle’s (2001) statement that GARCH(1,1) is the "simplest and most robust of the family of volatility models"9 . The seventh additional input required to price the CDS is the expected default boundary as a 7 8

The Markit database starts in January 2001, which limits our ‡exibility to de…ne the pre-SOX period. Although we do not have CDS spreads denominated in other currencies, we expect that they would be a¤ected

by SOX too. This is because many non-US …rms with CDS trading were cross-listed (equity) in the United States in 2002, and the main provisions of SOX applied to foreign registrants too. See Perino (2003), Litvak (2007), Li (2007), and Smith (2008) for discussions. 9 See pages 471-474 of Hull (2006). When the GARCH(1,1) estimation yields a non-stationary ("mean-‡eeing") model, which happens in 25 of the 252 sample …rms, we use we use a exponentially smoothed moving average of the previous 252 days with a smoothing coe¢ cient of 0.94.

7

fraction of reported liabilities, L: The CreditGrades Technical Manual (2002) suggests using the expected default boundary L =

1 2

for all …rms. We do this as a robustness check. In our base

results we choose a di¤erent L for each industry, chosen in order to maximize the total number of …rm-day observations in that industry in which market spreads are within the range of spreads 0

that can be delivered by the CreditGrades model for all values of . After …nding such L s, we calibrate

for each …rm-period so as to minimize the sum of squared di¤erences between market

and model CDS spreads. Appendix A provides additional details on the CreditGrades model and its calibration.

2.4

Data overview

Our sample includes 252 …rms, after merging the Markit database with CRSP and COMPUSTAT, excluding …nancial …rms and subsidiaries, and requiring at least 30 quotes per …rm in each period. Sample …rms are large: only 33 were not part of the S&P500 Index at some point in the sample period. Table 1 contains summary statistics for the spread and its determinants in the pre- and postSOX periods. The reported means and standard deviations are cross-sectional summary statistics based on …rm-speci…c time-series averages of the corresponding variable. In the table, one minus leverage is the stock price divided by the sum of the stock price and liabilities per share. Spreads are reported in basis points.

T ABLE 1 The mean spread is 119:3

111:2 = 8:1 basis points lower in the post-SOX period. As the Cred-

itGrade pricing formula shows, the CDS spread is a complex function of the model’s eight inputs. Thus, increased reliability in corporate reporting may not necessarily be the driver of the decrease in spreads following SOX. Equity volatility and risk-free rates decrease in the post-SOX period, which reduces credit spreads, holding other factors constant. However, average leverage increases and recovery rates decrease in the post-SOX period, which increases spreads, holding other factors constant. The mean number of time-series observations in the earlier period is lower than in the post-SOX period, while its standard deviation is higher. This is because the number of …rms in the Markit database has increased over time: not all 252 …rms in our sample were part of the Markit database as of January 1, 2001. Each …rm, however, has at least 30 observations in both the pre-SOX and post-SOX periods.

8

2.5

Hypotheses

Below we state the three hypotheses whose empirical validity we aim to assess. Throughout, corporate opacity refers to the uncertainty parameter

calibrated from market prices using the

CDS pricing model described earlier. Hypothesis 1: Corporate opacity is lower for …rms that have higher earnings quality, are perceived to be more transparent and to have better corporate governance. This can be seen as external validation of the corporate opacity parameter

: The calibrated

parameter presumably measures uncertainty about a …rm’s true leverage as perceived by investors. We expect this uncertainty to be inversely related to quantitative measures of earnings quality. We focus on three measures: accrual quality (Dechow and Dichev 2002; Ashbaugh-Skaife, Collins, Kinney, and LaFond 2008), abnormal accruals (Francis, Nanda, and Olsson 2008; Ashbaugh-Skaife et al. 2008), and earnings conservativism (Basu 1997; Zhang 2008). We also expect our calibrated opacity parameter to be inversely proportional to the number of independent directors in the …rm’s Board (Anderson, Mansi, and Reeb 2004). Moreover, calibrated opacity should be higher for younger …rms, which did not have enough time to build a reputation of reliable reporting, or have not yet "ironed out the kinks" in their internal control systems (Diamond 1989; Doyle, Ge and McVay 2007; Hyytinen and Pajarinen 2008). In addition to the aforementioned objective measures, we expect the the corporate opacity parameter to be negatively related to measures of corporate transparency based on expert judgement, such as the publicly available S&P Transparency and Disclosure Ratings of Pattel and Dallas (2002). It is important to point out that, for the purposes of this paper, our methodology remains valid even if the calibrated parameter

is a noisy measure of corporate opacity. Suppose that

is a

catch-all measure a¤ected not only by corporate opacity but also by other factors, such as model error or a …rm’s attractiveness for leveraged buy-outs. There is no ex ante reason to believe that model error should systematically change after the passage of SOX. Thus, assuming its impact is constant in the pre- and post-SOX periods, changes in

need be due to changes in corporate

opacity. Furthermore, leveraged buy-out activity increased substantially following (some would argue because of) the Act, which would act to increase rather than decrease credit spreads, and consequently our calibrated opacity parameter in the post-SOX period. Hypothesis 2: Corporate opacity decreases after the enactment of Sarbanes-Oxley. Existing research provides evidence that corporate reporting has become more reliable after SOX (Ashbaugh-Skaife et al. 2008; Cohen et al. 2008; Dyck et al. 2007; Hutton, Marcus, and Tehranian

9

2008; Singer and You 2008). Recent surveys con…rm research evidence. The majority of 274 …nance o¢ cers surveyed by the Financial Executives Research Foundation (2006) believe that SOX increased investors’ con…dence in …nancial reports. For large …rms, with more than $25 billion revenues, 83% of executives in the survey agree that investors are more con…dent in reported numbers as a result of SOX. Furthermore, 82% of audit committee members surveyed by the Center for Audit Quality (2008) think that audit quality has improved in recent years, while 65% of committee members believe that investors have more con…dence in capital markets as a result of SOX. Given the research and survey evidence, we argue that CDS market participants are less uncertain about the true level of corporate leverage following the Act. Thus, corporate transparency as perceived by investors has increased after SOX. Hypothesis 3: After the enactment of SOX, corporate opacity decreases more for …rms that are more likely to be a¤ ected by the Act. Firms whose reports are more reliable prior to SOX presumably already have better internal controls, more detailed disclosure, or more reliable auditing before the Act, which makes them less likely to be a¤ected by the new legislation. Consistent with this notion, Chhaochharia and Grinstein (2007) show that the net bene…ts of the new legislation are higher for …rms that are less compliant with the Act in the pre-SOX period. By the same logic, if the new regulation does indeed a¤ect corporate opacity, we expect its impact to vary with …rms’pre-SOX characteristics. Speci…cally, we predict that the decrease in opacity should be more pronounced for …rms that: are less transparent and have lower earnings quality in the pre-SOX period, and are more likely to be a¤ected by SOX according to the criteria in Chhaochharia and Grinstein (2007).

3

Empirical Analysis

As explained earlier, we calibrate a corporate opacity parameter before and after the enactment of SOX. We then use this measure to estimate the impact that a change in the reliability of corporate reporting has had on credit spreads. Figure 1 presents the time-series of the median observed spread and the median model-implied spread, calculated with the calibrated parameters. Model-implied spreads are based on …rm-speci…c parameters calibrated separately in the pre-SOX and post-SOX periods. There is a pronounced decrease in model spreads at the boundary between the pre-SOX and the post-SOX periods. This is consistent with the idea that the corporate opacity parameter may have decreased in the post-SOX period for the typical …rm in our sample. To determine if the model implied spreads decreased due to a decrease in opacity, however, we must control for the other determinants of credit spreads.

10

F IGU RE 1 There are cases in which the model over predicts spreads even when the opacity parameter

is zero.

This implies that, conditional on the …rm’s asset volatility and leverage, on the recovery value of debt, and on the level of the risk-free rate, model spreads are too high relative to observed spreads. See Panel B of Table A.1 for additional information on the corner solutions of the calibration procedure.

3.1

Testing hypotheses

In this section we discuss the empirical results of testing the three hypotheses presented earlier. Variable de…nitions are presented in Table 2.

T ABLE 2

3.1.1

Hypothesis 1

Table 3 presents means and medians of the pre-SOX opacity parameter

across …rms grouped by

characteristics related to …nancial reporting reliability. In each panel, we test the hypotheses that means and medians of corporate opacity are equal across …rms with high and low …nancial reporting reliability. The total number of observations in each grouping is below 252 when the corresponding characteristic is not available for …rms in our sample10 . Across all Panels, the break down between …rms with high or low …nancial reporting reliability is chosen so that sample sizes across bins are as similar as possible. Panels (A) through (C) of Table 3 are based on quantitative measures of earnings quality. Consistent with Hypothesis 1, the evidence shows that corporations with lower accrual quality, higher discretionary accruals, and less conservative earnings, tend to have higher calibrated opacity. Therefore, …rms with lower quality earnings tend to have higher cost of debt, 10

There are 39 …rms with virtually identical earnings conservativism, between 0.999 and 1.001. For these …rms,

the coe¢ cient on negative returns in the Basu (1997) regression was very close to 0. Since the median of earnings conservativism in the sample of 252 …rms was 1, and these …rms are less than 1/1,000 standard deviations apart from each other, keeping these …rms in the sample only adds noise. These …rms are dropped from the sample in Tables 3 and 5 (but not in Table 9). Had they been kept in the sample, sorting results would go in the same direction of those in Tables 3 and 5, but would be less strong statistically due to increased sampling noise.

11

ceteris paribus 11 . Results of Panels (D) and (E) show that, as conjectured, younger …rms and …rms with lower number of independent directors tend to have higher calibrated opacity. Panel (F) is based on Standard and Poor’s Transparency and Disclosure Index of June 2002, based on expert judgement rather than purely quantitative modeling. According to the measure, more transparent …rms tend to have lower corporate opacity.

T ABLE 3

3.1.2

Hypothesis 2

Table 4 Panel A reports statistics on the calibrated opacity parameter

in the pre- and post-SOX

periods. The post-SOX opacity parameters are less than or equal to the pre-SOX parameters on average and at each quartile. As shown by the scatter plot in Figure 2, most …rms in our sample experience a decrease in the calibrated opacity measure following SOX. The mass of calibrated parameters that equal zero corresponds to lower-bound solutions in the calibration (see Appendix A for details). Untabulated results show that the correlation between pre- and post-SOX opacity parameters is 0:78, while the Spearman rank-correlation is 0:79. The high correlations suggest that is associated with …rm’s intrinsic characteristics, rather than with noise in CDS spreads.

T ABLE 4 F igure 2 Table 4 Panel B provides a formal test of the hypothesis that pre- and post-SOX corporate opacity are drawn from distributions having the same mean or median. The mean (median) opacity parameter is 0:656 (0:510) in the pre-SOX period and decreases to 0:450 (0:391) following enactment of SOX, a 31% (23%) reduction. The di¤erences in means and medians across sub-periods are signi…cant at the 1% probability level, providing strong statistical support for the hypothesis that the distribution of the corporate opacity parameter shifts after SOX. However, it is di¢ cult to gauge the economic relevance of this evidence. Although a one quarter decrease in the opacity parameter appears to be substantial, its economic signi…cance needs to be assessed in light of its e¤ect on model-implied CDS spreads. In the next section, we provide a more detailed discussion of the economic signi…cance of the evidence discussed here. 11

Callen, Livnat, and Segal (2008) study the impact of earnings (not the reliability of reported earnings) on CDS

spreads.

12

3.1.3

Hypothesis 3

Figure 2 shows that, even though most …rms experience a decrease in calibrated opacity, there is substantial cross-sectional variation in the magnitude of the decrease. Our Hypothesis 3 is that the reduction in the opacity parameter following SOX is larger for …rms more likely to be a¤ected by the new legislation. Table 5 presents evidence consistent with Hypothesis 3. The table reports mean and median changes in the opacity measure for various subsamples obtained by grouping …rms based on pre-SOX characteristics. Consistent with Hypothesis 3, …rms with lower accrual quality, higher discretionary accruals, and less conservative earnings tend to experience a larger reduction of calibrated opacity following SOX. Moreover, …rms with lower number of independent directors, younger …rms, and …rms with poorer disclosure quality according to the S&P 2002 rating. also experience a more pronounced drop in opacity following SOX. Panel (G) is based on criteria adopted in Chhaochharia and Grinstein (2007). The authors argue that …rms with incidences of insider trading, restatements, and related party-transactions in the pre-SOX period should be more a¤ected by the passage of SOX because those events are manifestations of poor governance structures12 . Panel (G) of Table 5 show that such …rms (identi…ed by "Yes" in the table) display larger reductions in corporate opacity . The test statistics in the table suggest that di¤erences in the mean and median reduction in opacity are unlikely to be due to chance. Untabulated results show that the larger reduction in opacity for …rms ‡agged by Chhaochharia and Grinstein (2001) is not driven by one of the three criteria in particular: the decrease in opacity is larger for …rms grouped according to each of the three criteria.

T ABLE 5

3.2

Economic signi…cance

Our tests indicate that the CDS-calibrated opacity parameter

is related to the corporate reporting

reliability, and that pre-SOX levels of corporate reliability predict changes in

following SOX. In

this section, we estimate the change in the cost of debt due to increased reliability of corporate reports following SOX, the main economic question of the paper. In particular, we compute the di¤erence between model-implied spreads in the post-SOX period using post-SOX calibrated opacity parameters versus pre-SOX calibrated ones for each …rm-day in our sample. By keeping all the other seven inputs of the CDS pricing formula unchanged in the post-SOX period, and comparing 12

We are grateful to Vidhi Chhaoccharia and Yaniv Grinstein for generously providing us with their data. The

data includes a fourth dummy variable, audit services, which was zero for all …rms in our sample.

13

model-implied spreads calculated with pre-SOX and post-SOX calibrated opacity parameters, we are able to calculate the change in model-implied spreads that is only due to the reduction of corporate opacity. The average spread di¤erence across the 87; 663 …rm-day observations in the post-SOX period is

20:8 basis points per year, with a standard error of 3:1 basis points. This

standard error is heteroskedasticity robust and clustered by …rm. The median spread di¤erence is

17:7 basis points per year, with a standard error of 2:6 basis points. This standard error is

bootstrapped with clustering by …rm. Given that the median spread in the post-SOX period is 111 basis points, the implied decline in the cost of debt is economically substantial. To better gauge the economic consequences of the increased transparency perceived by investors following SOX, we compute the dollar savings that result from the implied decline in the cost of debt. We obtain from COMPUSTAT the total amount of (interest-bearing) debt for each …rm in our sample throughout the post-SOX period. We then multiply the spread di¤erence for each …rm on each day by the corresponding level of debt, and compute the time series median of this annual dollar saving value for each …rm. Taking the cross-sectional median of the time-series median annual dollar saving value, we estimate that the implied savings related to the cost of debt amount to $2:75 million per year for the typical …rm in our sample. Summing the median dollar savings across the 252 …rms, we estimate that the passage of SOX is associated with a total reduction in the cost of debt of $844 million per year for our sample …rms as a result of enhanced transparency.

4

Robustness checks

We perform two kinds of robustness checks in this section, . First, we explore the validity of other plausible explanations for the results presented earlier. Then, we assess the robustness of our main …ndings to changes in the way we calibrate the CDS pricing model.

4.1

Systematic risk: risk loadings and the price of risk

The CreditGrades model does not accommodate for di¤erences in CDS spreads due to di¤erences in systematic risk. It is possible that spreads are relatively higher for …rms in which the CDS spread contract value co-varies strongly with the overall state of the economy. Therefore, one could argue that the corporate opacity parameter

simply proxies for a premium for bearing systematic risk.

In the cross-section, we explore this possibility by comparing the calibrated ’s across subsamples of …rms that have di¤erent CAPM betas calculated with equity returns. We also measure bond systematic risk by calculating the loadings on corporate bond factors used by Fama and French

14

(1993) and Gebhardt, Hvidkjaer, and Swaminathan (2005). We measure returns by calculating CDS-implied corporate bond returns, as in Longsta¤ et al. (2007). The two corporate bond factors are calculated using Merrill Lynch bond index returns. The …rst, TERM, is the returns of a zeroinvestment portfolio long in long-term government bonds and short in T-Bills. The second, DEF (for default), is the return of a zero-investment portfolio long in BBB corporate bonds and short in AAA corporate bonds. Table 6 contains the results of this analysis.

T ABLE 6 The …rst two columns of Table 6 report test statistics for equality of means and medians of calibrated opacity across subsamples of …rms having high versus low risk loadings. There are six of such tests, encompassing three kinds of risk loadings and the pre- and post-SOX periods. Contrary to the risk-based explanation of our results, calibrated opacities are negatively related to loadings on the Market and the DEF risk factors in this sample. We argue that this result is driven by endogenous leverage: more opaque …rms, recognizing that they are charged relatively higher interest rates, choose to take on less debt. This makes them less prone to su¤er from liquidity shortages during recessions13 . It turns out that pre-SOX opacities are positively related to loadings in the TERM factor. However, the evidence is not robust to using medians rather than averages, which suggests that this result is driven mainly by outliers. Moreover, post-SOX opacities are not positively related to post-SOX loadings in the TERM factor. Since there is as much cross-sectional variation in the TERM factor before and after SOX, the …nding that opacities are related to loadings on the TERM factor is not robust. The combined results indicate that, to the extent that loadings on CAPM and Fama-French bond factors are good proxies for exposure to systematic risk, a systematic risk explanation of our results does not hold in the cross-section of opacity levels. The last column of Table 6 examines the possibility that the decrease in corporate opacity documented by us is actually picking up the e¤ect of a decrease in market-wide price of risk. For constant risk loadings, a decrease in the price of risk would have caused a decrease in systematic risk premia for all …rms, which we would capture in the form of lower opacity parameters. We test a cross-sectional implication of this explanation: since the risk premium is the product of an asset-speci…c risk loading by a market-wide price of risk, this alternative explanation implies that a 13

Acharya, Davydenko, and Strebulaev (2008) make a similar endogenous choice argument to explain their result

that credit spreads are positively associated to cash holdings. They argue that risky …rms choose to hold more cash which reduces the default risk in the short-run, but over the long-run their higher risk is re‡ected in higher credit spreads.

15

decrease in the market-wide price risk should a¤ect more …rms with high risk loadings. Therefore, …rms with higher pre-SOX risk loadings would display larger decrease in the opacity parameter. Similarly to the tests in levels of corporate opacity, results for the Market and DEF risk factors contradict the risk-based explanation. If anything, there was a larger reduction in opacity for …rms with low not high betas with respect to these factors. For the TERM risk factor, results are mixed: the mean test show that there was indeed a larger reduction in opacity for …rms with high loadings in the TERM factors, but the di¤erence in median opacity between …rms with high and low TERM betas is much smaller than the di¤erence in means, and not statistically signi…cant at 10%. Assuming that loadings in the Market, TERM and DEF factors indeed capture exposure to systematic risk, and that the price of risk of the TERM factor is not much greater than the prices of risk in the Market and DEF factors combined14 , these results suggest that it is unlikely that a reduction in the price of risk story is driving the bulk of our results. One may still argue that a market-wide price of risk explanation cannot be discarded because exposure to Market, TERM and DEF factors are bad proxies for exposure to systematic risk. To address this concern, we investigate the systematic risk explanation in the time series by calibrating the opacity parameter

for the 2004 and 2005 periods too. The cross-sectional average of

is 0:453

for 2004 and 0:474 for 2005. In both 2004 and 2005 the average parameters are close to the postSOX average of 0:450 and much lower than the pre-SOX average of 0:656, both reported in Table 4. The stability of the parameter across years in the post-SOX period provides little support to the alternative explanation based on time-varying price of risk.

4.2

Ratings and liability structure

The CreditGrades model does not di¤erentiate between types of liabilities nor does it incorporate non-public information about liabilities available to rating analysts and incorporated in credit ratings. Perhaps we feed the model an overly coarse measure of liabilities, while the market takes a much more nuanced look at the liability side of a …rm’s balance sheet. For example, while we ignore di¤erences between short- and long-term liabilities, these di¤erences may a¤ect CDS spreads and impact our calibrated opacity parameter. Analogously, since rating agencies have access to nonpublic information and incorporate such information in the rating process, it could be the case that CDS spreads re‡ect not only public balance sheet information but also non-public information conveyed by credit ratings. If that is the case, our measure of opacity could simply be proxying for the structure of a …rm’s liabilities and for the special information conveyed by ratings. We examine 14

Fama and French (1993) conclude that the price of risk of TERM (and DEF) factors is close to zero Gebhardt et

al (2005) …nd that the point estimates of the prices of risk of TERM and DEF are similar, but the DEF factor price of risk is statistically signi…cant, whereas the TERM factor price of risk is not.

16

this possibility by comparing our opacity parameters across subsamples segmented by credit ratings and the ratio between short and long term liabilities. The results are contained in table 7.

T ABLE 7 The evidence in Panel (B) of Table 7 shows that calibrated opacity is actually higher for …rms with higher credit ratings. This contradicts the argument outlined above. Therefore, calibrated opacity is not capturing information available in credit ratings,but missing from the balance sheet. The positive relationship between opacity and credit rating could be driven by an endogenous leverage e¤ect: more opaque …rms, recognizing that they are charge relatively higher interest rates on debt, choose to take on less leverage, and thus tend to have higher credit ratings. The evidence in Panel (A) suggests that the ratio of short-term to total liabilities may contaminate the calibrated measure of corporate opacity. The test statistics for the equality of means and medians show that, in the pre-SOX period, …rms with relatively more short maturity debt tend to have higher calibrated opacity. However, the evidence is weaker in the post-SOX period: the medians of opacity are very close for …rms with low and high ratios of short term to total liabilities. It may also be the case that more opaque …rm endogenously choose more shorter term debt, which could lead to a reduction of total debt costs over longer time periods. Finally, the last set of results in Table 7 are inconsistent with maturity composition of debt explaining away our results. In this test, we group …rms according to the change in the ratio of short-term to total liabilities from the pre- to the post-SOX periods, and calculate means and medians of the change in calibrated opacity. If the drop of calibrated opacity were explained by …rms lengthening the maturity of their liabilities, we would expect to see a larger reduction in opacity for …rms experiencing a larger decrease in the ratio of short term to total liabilities. Our tests, however, do not support this conjecture.

4.3

Supply of default insurance in the CDS market and the introduction of TRACE

The CreditGrades model does not accommodate potential demand and supply shifts in the market for default insurance that could a¤ect spreads if …nancial markets are not frictionless. For example, suppose CDS dealers had some degree of monopoly power in the pre-SOX period and they were net sellers of insurance in the CDS market, while being net buyers of insurance in equity and option markets (they should in principle be hedged overall). One could argue that dealers extracted rents from buyers of insurance by charging high spreads early in our sample period. If more dealers later entered the CDS market, this may have eroded the pre-SOX rents. In this case, the change in our 17

calibrated opacity parameter may be capturing the increased degree of competition in the CDS market in the post-SOX period. To examine this possibility, we use the average number of dealers providing daily quotes as a …rm-level proxy of the degree of competition in the CDS market in the pre- and post-SOX periods. If the decline in what we call opacity is due to increased competition in the CDS market, the decrease in opacity should be larger for …rms with a larger increase in the number of dealers providing quotes following SOX. The evidence in Panel (A) of Table 8 does not support this explanation. The equality of means test suggests that …rms with a larger increase in the number of brokers tended to have a smaller decrease in the opacity parameter. The equality of medians test indicates that di¤erences across subsamples are not statistically signi…cant. T ABLE 8

In addition to assuming a perfectly competitive market, the CreditGrades model does not incorporate market microstructure e¤ects that may in‡uence security prices. One could argue that the July 2002 introduction of TRACE in the corporate bond market and the associated increase in market transparency is responsible for the reduction of credit spreads (in excess of traditional spread determinants) we document. Goldstein, Hotchkiss, and Sirri (2007) provide some evidence that credit spreads decrease for bonds whose trading becomes more transparent with TRACE. It is possible that such reduction is transmitted to the CDS market by the CDS and bond arbitrage relationship (Du¢ e 1999). We investigate one cross-sectional implication of this alternative explanation. The introduction of TRACE was gradual and most …rms did not have bonds in the system until much later than July 2002. This allows us to test whether the e¤ect we document is at least partly driven by increased transparency in the bond market. We compute the fraction of the post-SOX period (August 2002 to December 2003) in which each …rm in our sample has bonds on TRACE, and label this fraction "time in TRACE". For example, time in TRACE is one for companies with bonds in TRACE since July 2002, 0:5 for companies with bonds …rst added to TRACE at the mid-point of the post-SOX period (April 2002), and 0 if no bonds were added by the end of 2003. Since our calibration uses spreads throughout the entire post-SOX period, the alternative explanation examined here implies that there should be a larger reduction in opacity for …rms with higher time in TRACE. The evidence in Panel (B) of Table 8 shows that this conjecture is not supported by data. The mean and median decreases in opacity are very close for high and low time in TRACE …rms.

18

4.4

Multiple regressions

The evidence presented so far is based on univariate sorting procedures. In this subsection, we address the possibility that some unforeseen interaction between potential explanatory variables may weaken our interpretation of the univariate results. In Panel A of Table 9, we report results of multiple regressions of the levels of pre-SOX calibrated opacity. Panel B contains results of multiple regressions of the changes in calibrated opacity from the pre- to the post-SOX periods. Explanatory variables are winsorized at 1% in each tail. In addition to the alternative explanatory variables described in the previous subsections, we include two control variables not explored in the paper before: volatility of stock returns and market capitalization. Volatility is an important control variable given Liu and Wysocki’s (2007) critique that pricing e¤ects attributed to accruals quality are due to innate business risk rather than accounting quality. Note that the sample size drops from 252 to 222 because 28 …rms not have all the accounting information needed to compute the accounting-based proxies of corporate reporting reliability15 .

T ABLE 9 Column (1) of Table 9 Panel A contains results of a Tobit Regression, while Column (2) contains results of a Censored Least Absolute Deviation (CLAD) Regression (Powell 1984; Chay and Powell 2001). These types of regressions are needed because the opacity parameter is bounded below by 0. With the exception of Discretionary Accruals, the coe¢ cients on …nancial reporting reliability variables have the correct sign in both regressions. F …rms with lower accrual quality, less conservative earnings, and younger …rms tend to have higher calibrated opacity. Coe¢ cients on these variables are statistically signi…cant at 5% in both the Tobit and the CLAD speci…cations. The coe¢ cients on Discretionary Accruals are statistically insigni…cant in both regressions. Note that the magnitude of the coe¢ cients on the …nancial reporting quality variables is reasonably close in both regressions, alleviating concerns that outliers could be driving the results. The coe¢ cients on stock return volatility and market capitalization are statistically insigni…cant. Panel B of Table 9 contains results of OLS regressions in Column (1) and Median regressions in Column (2). To interpret the sign of the coe¢ cients, recall that a negative sign means a larger decrease in the opacity parameter following SOX for larger values of the explanatory variable. In other words, negative signs are associated with "bad" characteristics from a cost of debt perspective. Con…rming the results in the levels speci…cation of Panel (A), the coe¢ cients on accrual quality, 15

We do not report results including the S&P Transparency and Disclosure Index or the Number of Independent

Directors as explanatory variables because their inclusion reduce the sample size from 222 to 174 and 175 respectively. Results are robust to inclusion of these variables.

19

…rm age, and earnings conservativism have the correct sign and are statistically signi…cant at the 5% level. Coe¢ cients on Discretionary Accruals and on Chhaochharia and Grinstein’s (2001) Dummy are not statistically signi…cant. Finally, as in Panel (A), the coe¢ cients on stock return volatility and market capitalization are statistically insigni…cant.

4.5

Changes in calibration procedure

In our baseline results we choose a di¤erent expected default boundary L for each industry using Fama and French’s 11-industry classi…cation, before we calibrate the corporate opacity

for each

…rm-period (see Appendix A for details). In Table B.1 we present an alternative calibration in which L is constrained to be equal to 0:77 across all industries. This is the single value of L across all …rms that maximizes the number of …rm-day spread observations falling within the boundaries of the CreditGrades model. All our results still hold. Table B.2 contains results of calibration using L =

1 2

for all …rms. This is the expected default boundary suggested by the CreditGrades

Technical Manual (2002). Most of our results hold in this alternative (worse) calibration. Finally, Table B.3 uses L’s chosen for each industry of our baseline results, but removing from the sample the …rms in which the calibrated opacity was a corner solution either on the pre-SOX or on the post-SOX periods. The total sample size drops from 252 to 156: The results are likely to be weaker than our baseline ones for two reasons. First, the sample size drops considerably, which reduces the power of our tests. Second, extreme …rms (very opaque or very transparent), likely to contain a lot of information about the relationship between spreads and opacity, are dropped from the sample. Nonetheless, Table B.3 shows that most of our results still hold in this subsample of interior solutions only.

5

Conclusion

Following a mounting number of high-pro…le corporate scandals, the US Congress passed the Sarbanes-Oxley Act in July 2002 in an attempt to restore public trust in US capital markets. The legislation aims to improve corporate transparency by altering governance, disclosure, internal control, and auditing practices of publicly traded companies. In this study, we investigate the impact of such changes on the cost of debt capital. In order to compute changes in the cost of debt capital due to changes in corporate transparency after SOX, we use daily CDS spreads and a structural CDS pricing model to calibrate a corporate opacity parameter for 252 …rms in each of the pre-SOX (January 2001 to July 2002) and post-

20

SOX (August 2002 to December 2003) time periods. First, we show that the opacity parameter is signi…cantly associated with …rm characteristics associated with the reliability of corporate reports. Firms with lower quality accruals, less conservative earnings, lower number of independent directors, lower S&P Transparency and Disclosure rating, and younger …rms tend to have higher CDS-calibrated opacity, and consequently higher cost of debt ceteris paribus. Second, we show that corporate opacity parameters tend to be signi…cantly lower in the post-SOX than in the pre-SOX period. Third, the decrease in the the opacity parameter tends to be larger for …rms more likely to be a¤ected by the new legislation. These …rms have lower accrual quality, less conservative earnings, lower number of independent directors, lower S&P Transparency and Disclosure rating; and are younger, and less compliant with SOX according to criteria in Chhaochharia and Grinstein (2007). We do not attempt to gauge our calibrated opacity parameter against alternative measures of corporate transparency or accounting/earnings quality, which may be a fruitful venue for future research. We argue, however, that our calibrated opacity parameter is uniquely well-suited to our goal of studying the e¤ect of changes in corporate transparency on the cost of debt. Our results indicate that the passage of SOX is associated with a substantial decline in the cost of debt due to increased reliability of corporate reports after SOX. We estimate that the reduction of opacity following SOX implies a 17:7 bp decrease in the 5-year CDS spread of the median …rm in our sample. Furthermore, we document that our results are robust to changes in our calibration procedure, and show the data does not support plausible alternative explanations for our …ndings.

21

References [1] Ait-Sahalia, Yacine, Per Mykland, and Lan Zhang (2005) "How Often to Sample a ContinuousTime Process in the Presence of Market Microstructure Noise". Review of Financial Studies 18, 351–416. [2] Acharya, Viral, Sergei Davydenko, and Ilya Strebulaev (2008) "Cash Holdings and Credit Spreads" Stanford University. [3] Anderson, Ronald C., Sattar A. Mansi and David M. Reeb (2004) "Board characteristics, accounting report integrity, and the cost of debt" Journal of Accounting and Economics 37, 315–342. [4] Ashbaugh-Skaife, H., D. W. Collins, W.R. Kinney, and R. LaFond (2008) “The e¤ect of internal control de…ciencies on …rm risk and cost of equity.” The Accounting Review 83, 217–250. [5] Ball, Ryan, Robert M. Bushman, and Florin Vasvari (2008) "The debt-contracting value of accounting information and loan syndicate structure" Journal of Accounting Research 46, 247–287. [6] Ball, Ray, Ashok Robin and Gil Sadka (2008). “Is …nancial reporting shaped by equity markets or debt markets? An international study of timeliness and conservativism”. Review of Accounting Studies 13, 168–205. [7] Bargeron, Leonce, Lehn, Kenneth and Zutter, Chad J. (2008). “Sarbanes-Oxley and Corporate Risk-Taking”. Working Paper. University of Pittsburgh. Available at SSRN. [8] Bartov, E., and D. A. Cohen (2007) “The Numbers Game in the pre- and post-Sarbanes-Oxley eras.” Working Paper. New York University. Available at SSRN. [9] Basu, S. (1997) "The conservativism principle and the asymmetric timeliness of earnings". Journal of Accounting and Economics 24, 3-37. [10] Black, Fischer and John Cox (1976) "Valuing corporate securities: some e¤ects of bond indenture provisions". Journal of Finance 31, 356–367. [11] Blanco, Roberto, Simon Brennan and Ian W. Marsh (2005) "An empirical analysis of the dynamic relation between investment-grade bonds and Credit Default swaps". Journal of Finance LX(5), 2255–2281. [12] Bushee, Brian J. and Christian Leuz (2005) "Economic consequences of SEC disclosure regulation: evidence from the OTC bulletin board" Journal of Accounting and Economics 39(2), 199–238. 22

[13] Callen, Je¤rey L., Joshua Livnat, and Dan Segal (2008) "The impact of earnings on the pricing of Credit Default Swaps" The Accounting Review, forthcoming. [14] Chay, Kenneth Y. and James L. Powell (2001) "Semiparametric Censored Regression Models" Journal of Economic Perspectives 15(4), 29-42. [15] Center for Audit Quality (2008) "Audit Committee Survey", March 2008. Available at http://www.thecaq.org/resources/caqresearch.htm. [16] Chhaochharia, Vidhi and Yaniv Grinstein (2007) “Corporate governance and …rm value: the impact of the 2002 governance rules” Journal of Finance 62, 1789-1825. [17] Chhaochharia, Vidhi and Yaniv Grinstein (2009) “CEO Compensation and Board Structure” Journal of Finance 64, 231-261. [18] Coates, John C. (2007) "The goals and promise of the Sarbanes-Oxley Act", Journal of Economic Perspectives 21(1), 91–116. [19] Collins, Daniel W. and Paul Hribar (2003) "Errors in estimating accruals: implications for empirical research" Journal of Accounting Research 40(1), 105–134. [20] Cohen, Daniel A., Aiyesha Dey and Thomas Z. Lys (2008) “Real and accrual-based earnings management in the pre- and post-Sarbanes Oxley periods”. The Accounting Review 83 (May), 757–787. [21] CreditGrades Technical Document (2002), http://www.creditgrades.com/resources/pdf/CGtechdoc.pdf [22] Cremers, K. J. M., Vinay B. Nair and C. Wei (2007) "Governance mechanisms and bond prices" Review of Financial Studies 20(5), 1359–1386. [23] Currie, Antony and Jennifer Morries (2002) "And now for capital structure arbitrage". Euromoney, December, 38–43. [24] Davydenko, Sergei A. and Ilya A. Strebulaev (2007) "Strategic actions and credit spreads: an empirical investigation". Journal of Finance 62, 2633–2671. [25] Dechow, P., R. Sloan, and A. Sweeney (1995) "Detecting earnings management" The Accounting Review 70: 193–225. [26] DeFond, Mark L., Mingyi Hung, Emre Karaoglu, and Jieying Zhang (2008) "Was the SarbanesOxley Act Good News for Corporate Bondholders?", University of Southern California, working paper.

23

[27] Demsetz, H. and K. Lehn (1985) "The structure of corporate ownership: causes and consequences" Journal of Political Economy 93, 1155–1177. [28] Diamond, Douglas (1989) "Reputation acquisition in debt markets" Journal of Political Economy 97, 828–863. [29] Doyle, Je¤rey, Weili Ge, and Sarah McVay (2007) "Determinants of weaknesses in internal control over …nancial reporting" Journal of Accounting and Economics 44, 193–223. [30] Du¢ e, Darrell (1999) "Credit swap valuation" Financial Analysts Journal 55(1), 73–87. [31] Du¢ e, Darrell and David Lando (2001) "Term structures of credit spreads with incomplete accounting information" Econometrica 69(3), 633–664. [32] Duarte, Je¤erson, Lance Young and Fan Yu (2008) "Credit spreads, governance and information asymmetry" University of Washington and Michigan State University working papers. [33] Duarte, Je¤erson, Francis Longsta¤ and Fan Yu (2007) "Risk and return in …xed Income arbitrage: nickels in front of a steamroller?" Review of Financial Studies, 20(3), 769–811. [34] Dyck, Alexander, Adair Morse, and Luigi Zingales (2006) "Who blows the whistle on corporate fraud"? Working paper. [35] Engle, Robert (2001) "GARCH 101: The Use of ARCH/GARCH Models in Applied Econometrics" Journal of Economic Perspectives 15, 157–168. [36] Fama, Eugene F. and Kenneth R. French (1993) "Common risk factors in the returns of stocks and bonds" Journal of Financial Economics 33, 3–56. [37] Fan, Hua, and Suresh M. Sundaresan (2000) "Default Valuation, Renegotiation, and Optimal Dividend Policy" Review of Financial Studies 13, 1057–1099. [38] Financial Executives Research Foundation (2006). "FEI Survey: Compliance Costs for Section 404", March. [39] Francis, Jennifer, Dhananjay Nanda and Per Olsson (2008) "Voluntary disclosure, earnings quality, and cost of capital" Journal of Accounting Research 46, 53–99. [40] Francis, Jennifer, Katherine Schipper, and L. Vincent (2005) "Earnings and dividends informativeness when cash ‡ow rights are separated from voting rights" Journal of Accounting and Economics 39, 329–360. [41] Francis, Jennifer, Ryan LaFond, Per M. Olsson, and Katherine Schipper (2004) "Costs of Equity and Earnings Attributes". The Accounting Review 79(4), 967–1010. 24

[42] Garlappi, Lorenzo, Tao Shu, and Hong Yan (2008) "Default Risk, Shareholder Advantage and Stock Returns". Review of Financial Studies 21, 2743–2778. [43] Gebhardt, William R., Soeren Hvidkjaer, and Bhaskaran Swaminathan (2005) "The crosssection of expected corporate bond returns: betas or characteristics?" Journal of Financial Economics 75, 85–114. [44] Goldstein, Michael, Edith Hotchkiss, and Eric Sirri (2007) "“Transparency and Liquidity: A Controlled Experiment on Corporate Bonds” Review of Financial Studies 20, 235–273. [45] Gompers, Paul A., Joy L. Ishii and Andrew Metrick (2003) "Corporate governance and equity prices" Quarterly Journal of Economics 118, 107–155. [46] Healy, Paul M. and Krishna G. Palepu (2003) "The fall of Enron" Journal of Economic Perspectives 17(2), 3–26. [47] Hermalin, Benjamin E. and Michael S. Weisbach (2007) "Transparency and corporate governance" NBER working paper n. 12875. [48] Hostak, Peter, Emre Karaoglu, Thomas Lys, and Yong Yang (2008) "An Examination of the Impact of the Sarbanes-Oxley Act on the Attractiveness of US Capital Markets for Foreign Firms". Working Paper. [49] Hyytinen, Ari and Mika Pajarinen (2008) "Opacity of young businesses: Evidence from rating disagreements" Journal of Banking & Finance 32, 1234–1241 [50] Hull, John C. (2006) “Options, Futures and Other Derivatives” 6th edition, Prentice-Hall. [51] Hull, John C., Mirela Predescu and Alan White (2004) "The relationship between credit default swap spreads, bond yields, and credit rating announcements" Journal of Banking and Finance, 28, 2789–2811. [52] Hutton, Amy, Alan J. Marcus and Hassan Tehranian (2008) "Opaque …nancial reports and the distribution of stock returns" Journal of Financial Economics, forthcoming. [53] Iliev, Peter (2007) "The E¤ect of the Sarbanes-Oxley Act (Section 404)" Brown University working paper. [54] Jones, J. (1991). "Earnings management during import relief investigation". Journal of Accounting Research 29: 193–228. [55] Jorion, Philippe, Charles Shi and Sanjian Zhang (2007) "Tightening credit standards: the role of accounting quality". Review of Accounting Studies, forthcoming. 25

[56] Jorion, Philippe and Sanjian Zhang (2007) "Good and bad credit contagion: Evidence from credit default swaps". Journal of Financial Economics 84 (3), 860–883. [57] Kang, Qiang, Qiao Liu and Rong Qi (2008) "The Sarbanes-Oxley Act and Corporate Investment: A Structural Assessment". University of Miami working paper. [58] Kothari, S.P., Andrew J. Leone and Charles E. Wasley (2005) "Performance matched discretionary accrual measures" Journal of Accounting and Economics 39(1), 163–197. [59] Leland, Hayne E. (1994) "Corporate debt value, bond covenants and optimal capital structure" Journal of Finance 53, 1213–1252. [60] Leuz, Christian (2007) "Was the Sarbanes-Oxley Act of 2002 really this costly? A discussion of evidence from event returns and going-private decisions". Journal of Accounting Economics 44, 146–165. [61] Leuz, Christian, Dhananjay Nanda, and Peter Wysocki (2003) "Earnings management and investor protection: an international comparison" Journal of Financial Economics 69, 505– 527. [62] Li, Xi (2007) "The Sarbanes-Oxley Act and Cross-Listed Foreign Private Issuers", University of Miami working paper. [63] Litvak, Kate (2007) "The E¤ect of the Sarbanes-Oxley Act on Non-US Companies Cross-Listed in the US". Journal of Corporate Finance 13, 195–228. [64] Liu, Michelle and Peter D. Wysocki (2007) "Operating risk, information risk, and cost of capital" MIT Sloan Working Paper. [65] Longsta¤, Francis, Sanjay Mittal and Eric Neis (2005) "Corporate yield spreads: default risk or liquidity? New evidence from the Credit Default Swap market". Journal of Finance LX(5), 2213–2253. [66] Longsta¤, Francis, Jun Pan, Lasse Pedersen, and Kenneth Singleton (2007) "How Sovereign is Sovereign Credit Risk?" UCLA working paper. [67] Mansi, Sattar A., William F. Maxwell, and Darius P. Miller (2004) "Does auditor quality and tenure matter for investors? Evidence from the bond market" Journal of Accounting Research 42, 755–793. [68] Molina, Carlos A. (2005) "Are …rms underleveraged? An examination of the e¤ect of leverage on default probabilities" Journal of Finance 60, 1427–1459.

26

[69] Patel, S.A. and George Dallas (2002) "Transparency and disclosure: overview of methodology and study results - United States" Standard&Poor’s report [70] Perino, Michael A. (2003) "American corporate reform abroad: Sarbanes-Oxley and the foreign private issuer". European Business Organization Law Review 4, 213–244. [71] Pittman, Je¤rey A. and Steve Fortin (2004) "Auditor choice and the cost of debt capital for newly public …rms" Journal of Accounting and Economics 37, 113–136. [72] Powell, James L. (1984) "Least Absolute Deviations Estimation for the Censored Regression Model" Journal of Econometrics 25, 303–325. [73] Schaefer, Stephen M. and Ilya A. Strebulaev (2008) "Structural models of credit risk are useful: evidence from hedge ratios". Journal of Financial Economics 90(1), 1–19. [74] Sengupta, P. (1998) "Corporate disclosure quality and the cost of debt". The Accounting Review 73(4), 459–474. [75] Singer, Zvi and Haifeng You (2008) "The E¤ect of Section 404 of the Sarbanes-Oxley Act on Financial Reporting Quality" McGill University working paper. [76] Smith, Geo¤rey P. (2008) "What are the e¤ects of SOX on cross-listed companies?", University of Illinois at Urbana-Champaign working paper. [77] Teoh, Siew Hong, and T.J. Wong (1993) "Perceived Auditor Quality and the Earnings Response Coe¢ cient", The Accounting Review 68(2), 346–366. [78] Verrecchia, Robert (1983) "Discretionary Disclosure" Journal of Accounting and Economics 5, 179–194. [79] Wittenberg-Moerman, Regina (2008) "The role of information asymmetry and …nancial reporting quality in debt trading: evidence from the secondary loan market", Journal of Accounting and Economics, forthcoming. [80] Yu, Fan (2006) "How pro…table is capital structure arbitrage?" Financial Analysts Journal, 62(5), 47–62. [81] Yu, Fan (2005) “Accounting transparency and the term structure of credit spreads". Journal of Financial Economics, 75(1), 53–84. [82] Zhang, Ivy Xiying (2007) "Economic consequences of the Sarbanes-Oxley Act of 2002" Journal of Accounting Economics 44, 74–115.

27

[83] Zhang, Jieying (2008) "The contracting bene…ts of accounting conservativism to lenders and borrowers", Journal of Accounting and Economics 45, 27–54. [84] Zhang, Benjamin Yi-Bin, Hao Zhou, and Haibin Zhu (2008) "Explaining Credit Default Swap spreads with the equity volatility and jump risks of individual …rms" Review of Financial Studies, forthcoming.

28

Appendix A: Model and Calibration Details The CreditGrades model (2002) is an adaptation and extension of the Black and Cox (1976) debt pricing model. Total …rm value per equity share is a Geometric Brownian Motion with zero drift and volatility : Reported liabilities per equity share is constant at D. Default happens the …rst time the value process hits an uncertain default boundary given by LD, where L is lognormally distributed and independent of the value process Vt . The expected value of L is L, and the standard deviation of the log of L is : Note that L can be below 1: structural models with endogenous default (e.g. Leland 1994) show that equity holders may be willing to keep the …rm alive even when the current value of assets is below the face value of debt. If the …rm defaults before the expiration of the CDS contract, the seller of protection stops receiving spread payments and has o make a lump-sum payment pay of (1 R). Given these assumptions, the CreditGrades manual (2002) shows that the fair CDS spread is well approximated by the closed-form formula in Section 2.1. It is important to mention that the credit spread is not a monotonic function of the uncertainty parameter . Given the other seven inputs of the CDS pricing formula, there is a

such that the

function c(T; ) reaches a maximum spread. This is the only critical point of the function c(T; ): the function is monotonically increasing for 0